With the advent of Internet and resulting data boom, Recommender Systems have come to rescue by filtering the information available on the internet by providing us with relevant information. These systems come handy when one wants to listen to songs, watch movies or even buy products on the Internet. Primarily, these recommender systems used content based or collaborative filtering techniques to recommend items. More recent research has studied the importance of contextual features in recommender systems. Music preference has always been associated with the contextual feature emotion. However, few studies study the mood congruence effect in the domain of music recommender systems. The field of music emotion recognition also remains unexplored with recommendations being made with limited features. \\

This master thesis analyses the relationship between few latent musical features and user emotion through our interface MooDify. It is a music recommendation system that incorporates emotion in a user using emotion induction techniques and investigate the effect of their emotional state on satisfaction and unexpectedness when presented with songs curated to specific musical features. To achieve this, we analysed the enjoyment and unexpectedness ratings for recommendations specific to latent musical features for a given emotional state. We have been able to achieve some interesting results through this study which has been discussed later in this work.